Approximate dynamic programming with recursive least-squares temporal difference learning for adaptive traffic signal control

In this study, an approximate dynamic programming approach with function approximation is applied to the scheduling of adaptive traffic signal control at isolated intersection. By using the linear function approximation, parameter adjustment is determined by the recursive least-squares temporal difference learning. The traffic modeling is based on the framework of Markov decision process. The proposed method can tackle the problem in the curse of dimensionality caused by the large state-action space in traffic model, especially in the adaptive control mode suggested in this paper. By comparing with other traffic control methods, the simulation results show that, our proposed algorithm can perform efficiently and quite well in real-time operation.